Providers of mapping-related services are continually challenged to accurately map and represent features and objects (e.g., buildings, trees, cars, other structures, etc.) in a mapping database. For example, one area of development has been in three-dimensional mapping, which historically has been dependent on manual effort for generating accurate models (e.g., three-dimensional models) of map features or objects. However, because of the vast numbers of objects that may be present in a mapped geographical area, manually creating models for each object or feature can be a significant resource burden. To reduce this burden, service providers can use, for instance, aerial or satellite photogrammetry to create surface models of a large geographical area (e.g., an entire city) from which individual object models can be extracted. Extracting models in this way, however, can result in models that include additional artifacts beyond the object being modeled (e.g., an extracted building object may include both the building and an adjacent tree, car, or other nearby structure). Accordingly, service providers face significant technical challenges for efficiently removing these artifacts from object models.